Question i5 6600k "spiky" temps and wattage during light tasks (with logs!)

[supaluminus]

This is my first overclock. I've been trying to push my little i5 as far as it will go without crashing or generating errors during benchmarks.

So far, I've managed to squeeze the chip for 4.50 GHz at 1.47v. This is just from adjusting the core multiplier and vcore. I don't know anything about load-line calibration, but I'm interested in learning more about it, especially if it will help me sustain a stable overclock.

Under load, temps have remained below a maximum of 90 C and an average of around 80 C. For benchmarks, I've been running OCCT Linpack 2021, Prime95, and Cinebench for around 15 minutes each for every step up in frequency and/or voltage.

Counting each increment, this has been very time consuming, but I've been enjoying the learning process, and I hope to learn a little more from experienced overclockers by sharing this post.

My concern at the moment has to do with temps and power consumption while performing light tasks.

The below chart features a comparison between stock settings (green) and my overclock (red) while watching a stream on Discord with a few browser tabs minimized in the background. I ran each log for around one hour.

https://imgur.com/xWqc86L

3.50 GHz 1.2vcore (green) vs. 4.50 GHz 1.47vcore (red)

PLEASE NOTE: The above CPU overclock includes a memory overclock of 3.20GHz at 1.35v. Stock CPU metrics were logged with stock memory frequency and voltage.

Average temp and power consumption increased from stock to overclock as expected, but I'm concerned about how "spiky" this wave is. In short: should I be concerned about this?

From what research I have done, large swings in wattage and temperature like this aren't good for the chip. It suggests to me that my overclock is unstable, but I really don't feel like I know what I'm talking about.

If I'm wrong, then what can I glean from this information? If I'm right, is there anything I can do, other than stepping down my multiplier and vcore, to stabilize my overclock? Is this when I should be thinking about load-line calibration, or would that make no difference?

Thanks for your time. I can include additional charts if necessary, as well as provide links to the log files themselves.

EDIT: CPU cooler and motherboard provided here, as requested by a moderator.

CPU Cooler: Kraken x61
Motherboard: GA-Z170X Ultra Gaming rev 1.0

 

[Shmee]

First, what are your full system specs? Specifically motherboard and CPU cooler are important to know. That said, 1.47 Vcore does seem too high. There are likely other settings in your UEFI setup that may help you get a more stable OC with lower Vcore, LLC being one of them to try.

 

[supaluminus]

First, what are your full system specs? Specifically motherboard and CPU cooler are important to know. That said, 1.47 Vcore does seem too high. There are likely other settings in your UEFI setup that may help you get a more stable OC with lower Vcore, LLC being one of them to try.

I tried responding to this and the site keeps throwing up a message refusing to post my reply because it looks like spam.

Here's the bare minimum. I don't know what's triggering the spam filter.

CPU Cooler: Kraken x61
Motherboard: GA-Z170X Ultra Gaming rev 1.0

So the variance I'm seeing in power draw and temps is a sign of instability then, right? It shouldn't be all over the place like that, ideally. Right?

 

[Shmee]

I don't think I would say it is necessarily unstable, it seems like you have stress stressed the CPU pretty thoroughly, though it could be in the future due to possible degradation. The variance could be somewhat by design due to larger differences in high vcore and lower vcore between load and ~idle workloads. This might also depend on what settings you are tweaking with regards to power savings, if at all. I hope this kind of makes sense.

My biggest concern would be that the Vcore is set too high, I would stay under 1.4V, or even 1.35V preferably. I never messed with OCing Skylake CPUs myself, but I know there is similar guidance for Haswell and Ivy Bridge. Skylake may be even more sensitive to high voltages.

My next suggestion would be to look into Z170 OC guides, for more information on fine tuning settings on your board for overclocking.

 

[coercitiv]

Based on what I remember about Skylake, going over 1.4V is not a good idea for long term use. I would go down at 1.35V static voltage, find the highest stable frequency under load, then push a bit more with a somewhat more aggressive LLC setting. Unless the chip is bellow average quality, you should be able to come close to 4.5Ghz at much lower voltage than your current 1.47V - a quick primer on LLC and core voltage can be found here.

Your idle temps can improve if you enable sleep states in BIOS. I know OC guides say to disable them for max oc potential, but you can always give up 100Mhz for much lower temps. This will bring power usage down greatly under light load and will also indirectly lower temp spikes. Speaking of temp spikes, you may want to check your minimum water pump speed, the water pump will be the limiting factor in absorbing short bursts of heat.

So the variance I'm seeing in power draw and temps is a sign of instability then, right? It shouldn't be all over the place like that, ideally. Right?

Nope, that's expected behavior. With the voltage and frequency you're pushing right now, your CPU is using 80%+ more power. Your captures are not identical either, so the sipkes in temperature are also coming form a difference in workload (whatever tasks were running at the time). Lowering voltage down to 1.35V should be a massive help, the same applies for increasing water pump speed if it's currently set to work at low speed during light loads.

Last bu not least, calibrate your final overclock without the help of air conditioning in the room (or using lower fan speed than normal if you dont' have a warm environment). Overclock stability with a static voltage also depends on max CPU temperature, so it's important to make sure there's headroom available here, not have the best cooling while stress testing to check final stability.

 

[BonzaiDuck]

Yes . . . Surprising that the OP just started this thread, when the SKY and KABY processors are vintage 2016 or 2017 technology. I've got three of these: a Skylake i7-6700K, Kaby Lake i7-7700K, and a 6700 locked processor. I had successfully overclocked the K processors until I had a static charge accident in 2021 with my original prototype system which blew the USB controller -- KEEP YOUR VAPING PENS AWAY FROM USB PORTS on dry, Santa Ana windy days or ground yourself on wool carpet! GETTING HIGH WITH GUMMY BEARS IS SAFER! After I refurbed that system, I had acquired spare parts and built a twin with the Z170 WS workstation board, but I never OC'd those systems. I got good performance anyway -- given my formula for building those rigs.

Sadly, for swapping out the Z170 motherboard on the damaged system, I lost all of my OC profiles stored in BIOS, or I could offer some insight here.

I WILL say THIS, though . . . I remember researching the VCORE settings for these, and I believe Intel had informed their "Board Partners" that 1.4V was a safe limit for default "AUTO" board settings. If I remember, I was able to get very stable overclocks at 4.7 Ghz with (maybe) 1.38V. And --yes-- an LLC setting is part of the formula. Ultimately you would want to use a program like IntelBurnTest, which shows iterative FFT results, so you can use simple spreadsheet tools to analyze the standard deviation or variance so you don't leave the CPU with insufficient voltage to eliminate errors which occur without crashing or terminating your test program. One of our former veteran members -- "IDontCare" -- had observed this phenomenon of the CPU recovering from errors and continuing to test with more variability in the results.

After the vaping pen accident, trouble-shooting and rebuild, I had to conclude that increasing the socketed RAM (3200 -- 14, 14, 14 XMP) from 32GB to 64GB and caching even my NVME drives to RAM with PrimoCache was all I needed to do with these Bad Boys . . . .

When I built these systems, I did it in high style for the cooling solution: I sent both K processors off to "Silicon Lottery" in Texas for delid-relidding and Grizzly Conductonaut liquid metal for $50 each. I don't think SL is in bidnis anymore . . . .

Once again, I really want to extend my use of Windows 10. These systems were the best I ever built.

 

[maddogmcgee]

Using stress tests like linpack and prime on 1.47 is not going to be good. Max voltage went down with each gen and 4.7 was well above recommended even for Sandy. Overclock also went way down as Intel started leaving less headroom at stock. I guess I would ask what the point is. I have a bunch of 6600 aios and they are great for word and chrome but I can't imagine a small overlook suddenly making them suitable for high end computing. If money is an issue, keeping the working computer working is probably a wise decision.